frorein Science (1995), 4:2487-2498. Cambridge University Press. Printed Copyright 0 1995 The Protein Society ~
~~
__
__
-
~~~~~
in the USA.
~~~~~~~~~
Solution structure of the catalytic domain of human stromelysin complexed with a hydrophobic inhibitor
STEVEN R. VAN DOREN,]s2 ALEXANDER V. KUROCHKIN,' WEIDONG HU,' QI-ZHUANG YE,3 LINDA L. JOHNSON,3 DONALD J. HUPE,3 AND ERIK R.P. ZUIDERWEG1s4
' Biophysics Research Division,
The University of Michigan, Ann Arbor, Michigan 48109-1055 of Biochemistry, University of Missouri-Columbia, Columbia, Missouri 6521 1 Department of Biochemistry, Parke-Davis Pharmaceutical Research, Division of Warner-Lambert Company, Ann Arbor, Michigan 48105 Department of Biological Chemistry, The University of Michigan Medical School, Ann Arbor, Michigan 48109-0606
* Department
(RECEIVED
July 27, 1995; ACCEPTEDSeptember 25, 1995)
Abstract Stromelysin, a representative matrix metalloproteinase and target of drug development efforts, plays a prominent role in the pathological proteolysis associated with arthritis and secondarily in that of cancer metastasis and invasion. To provide a structural template to aid the development of therapeutic inhibitors, we have determined a medium-resolution structure of a 20-kDa complex of human stromelysin's catalytic domain with a hydrophobic peptidic inhibitor using multinuclear, multidimensional NMR spectroscopy. This domain of thiszinc hydrolase contains a mixed 0-sheet comprising one antiparallel strand and four parallel strands, three helices, and a of 20 structures was calculated using, on avmethionine-containing turn near the catalytic center. The ensemble 166-residue protein fragment complexed with a 4-residue erage, 8 interresidue NOE restraints per residue for the substrate analogue. The mean RMS deviation (RMSD) to the average structure for backbone heavy atoms is 0.91 A and for all heavy atoms is 1.42 A. The structure has good stereochemical properties, includingits backbone torsion angles. The&sheet and a-helicesof the catalytic domainsof human stromelysin (NMR model) and human fibroblast collagenase (X-ray crystallographic modelof Lovejoy B et al., 1994b, Biochemisfry 33:82078217) superimpose well, having a pairwise RMSD for backboneheavy atoms of2.28 A when three loop segments are disregarded. The hydroxamate-substituted inhibitor binds across the hydrophobic activesite of stromelysin in an extended conformation. The first hydrophobic side chain is deeply buried in the principal Si subsite, the second hydrophobic side chain is located on the opposite side of the inhibitor backbone in the hydrophobic Si surface subsite, and a third hydrophobic side chain ( P i ) lies at the surface. Keywords: catalytic domain; hydroxamate; matrix metalloproteinase 3; multidimensional NMR; protein structure; stromelysin
As a member of the zinc- and calcium-dependentfamily of matrix metalloproteinases (MMPs),which hydrolyze the extracellular matrix, stromelysin participates in the tissue remodeling of health and disease. Regarding the shared domain structure of the MMPs,their subfamilies, specificity, transcriptional regulation, activation, and posttranslational regulation,see reviews of Woessner (1991), Matrisian (1992), Birkedal-Hansen et al. (1993), and Nagase (1995). The involvement of MMPs in tissue remodeling includes that of development, tissue resorption, reproduction, angiogenesis, and wound healing, and that of sev-
era1 diseases including arthritis and cancer. Theinvasive potential of human tumors is correlated with the expression of several MMPs (Stetler-Stevenson et al., 1993) but the particular MMPs expressed vary among types of tumors. Highly elevated levels of stromelysin (MMP-3), alongside other MMPs, have been measured in many carcinomas of the breast, bone, and esophagus (Clavel et al., 1992; Sasaguri et al., 1992; Shima et al., 1992). MMP-3 plays a principal role injoint destruction aswell. The concentration and proteolyticactivity of stromelysin have been found to be elevated dramatically in joints afflicted by either osteoarthritis (Okada et al., 1992; Wilhelm et al., 1993) or by rheumatoid arthritis (Okadaet al., 1989; Gravallese et al., Reprint requests to: Erik R.P. Zuiderweg, Biophysics Research Division, The Universityof Michigan, 930 North University Avenue, Ann 1991; Firestein & Paine, 1992; Walakovits et al., 1992). ConseArbor, Michigan 48109-1055; e-mail:
[email protected]. quently, stromelysin and other MMPs are drug targets for which 2487
2488
S.R. Van Doren et al.
weg,1988; Marion et al., 1989). Inparticular,'5N-resolved atomic resolution structures have been sought to aid the drug design process. NOESY (as described by Van Doren et al. [1993]) and "CA representative of the MMPs, the catalytic domain of huresolved NOESY (as modifiedby Majumdar & Zuiderweg [ 19931) man stromelysin was shown by NMR to contain three helices spectra acquired in H 2 0 using gradient enhancements and mixand a five-stranded mixed 0-sheet (Gooley et al., 1993; Van ing times of 70 ms were used. Protein-protein NOEassignments Doren et al., 1993), organized in a fold similar to that of crayalso relied upon a 3D"C-resolved NOESY (as modified by fish astacin (Bode et al., 1992). The zinc-binding motif at the Van Doren & Zuiderweg [1994]) collected in D 2 0 with a mixing time of 55 ms. Assignments of the peptide-derived inhibiactive site, containing the conserved HEXXHXXGXXH sequence, aswell as a conserved methionine in a turnadjacent to the tor relied on a TOCSY (Bax et al., 1994) and on a NOESY catalytic zinc, are also sharedwith astacin. These structural fea(Ikura & Bax, 1992), which suppress virtuallyall '3C-bound tures, shared not only among thematrix metalloproteinases and proton signals from the uniformly '3C/'5N-labeled stromelysin astacins, but also among theserralysins and adamalysins,allow in order to reveal the I2C-bound protons of the unlabeled, them to be grouped into a superfamily dubbed the metzincins hydroxamate-substituted peptide-like inhibitor. Theinhibitor as(Bode et al., 1993). In efforts to aid drug design, crystallographic signments were supplemented by a "C/"N-filtered NOESY structures of the catalytic domain of fibroblast collagenase or (modified from Ikura & Bax [1992]) of the same sample in MMP-1 (Borkakoti et al., 1994; Lovejoy et al., 1994a, 1994b) H 2 0 . These NOESY spectra,suppressingthe I3C- and "Net al., 1994; and of neutrophil collagenase or MMP-8 (Bode bound protons of the protein (NOESYmixing times of 100 ms), Reinemer et al., 1994; Stams et al., 1994) have become available. were used in conjunction with the previously mentioned 3D NOESY spectra of shorter mixing times for identifying proteinThe structure of full-length porcine MMP-I shows that the C-terminal hemopexin-like domain is a four-bladed ''0propelinhibitor NOES. ler" (Li et al., 1995). An NMR solution structure of stromelysin catalytic domain at pH5.5 complexed with an N-carboxy alkyl The fold inhibitor, N-(R-carboxy-ethyl)-a-(S)-(2-phenylethyl)-glycyl-LThe amino-terminus of mature stromelysin lies along the third arginine-N-phenyl amide, has been published as well (Gooley et al., 1994), and has since been further refined (Protein Data helix (C) of the catalytic domain and leads into &strand I (the Bank access code 2SRT). secondary structure elements are defined in the legend to Fig. I ; We present here the solution structureof human stromelysin's strands are shownas flat strips in Fig. 1 and in yellow in Fig. 2; catalytic domain under conditions of higher calcium affinity Kinemage 2) of the mixed 0-pleated sheet. Strand 1 is terminated (pH 7 . 0 ) , and concentration, complexed with a hydroxamate by a bulge that precedes an extended segment of a loop, paralinhibitor of more hydrophobic character ((R,S)-N-[2-[2-(hylel to the sheet but not hydrogen bonded to the sheet. The loop droxamino)-2-oxoethyI]-4-methyl-l-oxopentyl]-~-leucyl-~turns back into the long and perfectly amphipathic helix A phenylalaninamide; DiPasquale et al., 1986) and in the presence (shown in white in Fig. 2). A short loop then turnsback immeof acetonitrile as a co-solvent. Our solution structureof the catdiately to connect to 0-strand11 at the edgeof the sheet. Thus, alytic domain of human stromelysin reveals the binding mode parallel strands I and 11 and helix A form a classic right-handed of this peptide substrate analogue in the hydrophobic pockets P-ol-6 motif. From the end of strand 11, a loop crosses over of the active site (Kinemages 1 and 2) (Protein Data Bank acstrand I on theconvex side of the sheet leading to strand111 runcess codes lUMT and IUMS). ning parallel to and between stands I and V. From strands 111 to IV, an S-shaped loop traverses the convexside of the sheet. The first or upper lobe of the S shares in coordinating the strucResults tural zinc ion (white sphere in Fig. 2), whereas the second or Assignments lower lobe binds a structuralcalcium ion (blue spherein Fig. 2; Kinemage 2). Following this loop at the edge of the 0-sheet, The backbone assignments and secondary structure were previstrand IV runs (antiparallel to strand V) along the active site cleft ously reported by Van Doren et al. (1993). The bulk of ' H a n d 13C it borders. Strand V, sandwiched between strands I11 and IV, side-chain assignments were obtained using cross-polarizationprovides a histidyl nitrogen ligand to the structuralzinc ion as driven HCCH-TOCSY spectra (Majumdaret al., 1993). CBCA does strand IV. Altogether, the sheet is amphipathic with the ex(C0)NH (Grzesiek & Bax, 1992) and HBHA(C0)NH (Grzesiek posed regions of the convex side beinghydrophilic and the con& Bax, 1993) experiments proved helpful in correlating backcave interior being hydrophobic. Strands 11, I , 111, and V run bone and side-chain groups. Due to short T,s in general for this parallel in that order. protein and for the methylene groups in particular, assignment A portion of the loop joining strandV with helix B borders of methylene groups was problematic, but a 3D HCH (Yamathe active site. The central helix B is mostly buried andcrosses zaki et al., 1993) with semi-constant time enhancementsproved very helpful in such cases. For the most difficult of methylene helix A with an angle of about 42". Helix B lines the bottom of assignments and for connecting aromatic ring systems, "N- the active site cleft. The C-terminal end ofhelix B contributes two histidines, which coordinate the Zn involved in catalysis and i3C-resolved NOESY spectra were employed. The bulk of (white ball in Fig. 2; Kinemage2). The third histidine ligand to the assignments, however, were obtained from scalar correlations. The assignments are currently more than 90% complete.this zinc is positioned a few residues away in the long loop connecting the Bhelix with the C helix. Met 219, conserved throughStereospecific valine and leucine methylassignments of the out the greater family of "metzincin" zinc endoproteases, lies protein, but not inhibitor, were obtained from a ["CIHSQC adjacent to the active site histidines. A few residues from this spectrum of a 10% uniformly "C-labeled sample (Neri et al., methionine is an extended sequence of hydrophobic residues that 1989). Protein-protein NOE assignments were obtained from forms the wall of the active site cleft opposite the wall of the cleft 3D heteronuclear-resolved NOESY experiments (Fesik & Zuider-
2489
Fig. I . Topology of the stromclysin catalytic domain. Flattened arrows represent the &strands and wound arrows represent e-helices. A ball-and-stick plot represents the heavy atoms of the inhibitor (IC1 U24522: DiPasquale et al., 1986) comprised of a hydroxamate residue followed by an isobutyl-containing residue, a leucine, and a phenylalaninamide. (Systematic name: (K.S)-N-[2-[2-(hydroxamino)-2-oxocthyl]-4-methyl-l-o~opcntyl]-~-l eucyl-L-phenylalaninamide.)The catalytic zinc ionis represented by a white sphere, carbon atomsby black spheres, nitrogen atoms by medium gray spheres, and oxygen atomsby light gray spheres. The average structure subjectedto minimization with restraints was plotted using MOLSCRIPT (Kraulis,1991). The first and last residues of each of the five d-strands ( I - V ) and three tu-helices (A-C) are shown as follows: I (H96-1101); I I (T131-RI34); I l l (1142-A147): IV (AI65-YI6X); V (G176-DIXl); A (K110-VI27); R (1.195-S206); and C (Q236-Y246).
formed by strand IV. The loop continues to where i l makes hydrophobic contacts with helix A. Ser 235 likely serves as thc N residue and D238 as the N 3 residue of an N-capping box of helix C (having reciprocal side-chain to backbone amide hydrogen bonds), judging from the characteristic upfield shifted I3Ca and downfield-shifted '.'C@resonances as described by Gronenborn and Clore(1994) to be indicative of such structures. Helix C lies along the surface of the domain at an angle of roughly 80" relative to helix A. Helix C has a few breaks in medium range NOEs and in protection of amides from exchange (i.e., of G241 and S244) and some irregularity of backbone torsion angles. The C-terminal eight residues of the catalytic domain studied are disordered and have been omitted from calculations of the ensemble.
+
Precision
The mean RMS deviation (RMSD) among the ensembleof 20 structures to the average structure for backbone heavy atoms is 0.91 f 0.06 A and for all heavy atoms is 1.42 5 0.06 A (see Kinemage I). Regions of lowest RMSD (Fig. 3) are correlated with the regions for which the most medium- and long-range NOE (histogram of Fig. 4; Table 1) constraints are available. The number of such NOE distance restraintsis correlated with the occurrence of regular secondary structure andwith the degree of burial in the hydrophobic core. Most of the regions of regular secondary structure have backbone RMSDs (from the average structure) ofless than 0.6 A. The helices can be identified in Figure 4 by the presence of medium-range NOEs. Angular order parameter statistics (Fig. 5; Hyberts et al., 1992) suggest most backbone torsions arewell defined to the extent that the@ and $ order parameters exceed 0.8, a threshold reflecting a standard deviation of less than 38", for thegiven dihedral angle throughout the ensemble. Portions of four loops in the model have multiple residues with order parameters below this threshold: amino-terminal residues 83-94; residues 147-
156 of the Ill-IV metal-binding loop; residues 188-192 of the V-B loop; and residues 208-232 of the B-C loop. The higher positional RMSDs are found in these regions as well. These regions are correlated with surface exposure and higher amide ex-
Table I . Surnrnary of restraints ~rsed in stnrctlrre calculations
Kcstraint type Assigned NOE crosspeaks used in calculations" Total NOE distance restraints Total protein-protein NOE distance restraints Intrarcsitlue Sequential ( t i - j l = I ) Medium range (2 II i - j l I4) Long range ( 1 i - j I > 4) Total inhibitor NOE distance restraints Sequential and I i - j l = 2 Long range inhibitor-protein Mean deviation above all NOE upper bounds, A Mean largest NOE violation, A I>ihetlral angle (6)restraint\ Hydrogen bond distance restraints (2 per H-bond) Total metal-ligand distance restraints Zinc-nitrogen Zinc-oxygen Calcium-oxygen
Number of restraints 1.386 1.336 I .281 3 510 302 466 55 3 52 0.037 5 0.001 0.64 0.07 55 x4 15 6 3
*
6
'' In 50 cases, NOES were present to both geminal protons of a methylene group or to both methyl groups of a side chain, but only one (rather than two) restraint could be used. That is, due to a software limitation, a single restraint was defined to the pseudoatom defined for the given group rather than properlyto both constituents of the pseudoatom. This accounts for the diflerenceof 50 between the numberof assigned crosspeaks and NOE restraints used.
S.R. Van Doren et al.
2490
P
Fig. 2. A: Ensemble of the 20 accepted structures of the catalytic domain of stromelysin representedby their C a traces. Heavy atom5 of the bound substrate analogue, apart from the hydroxamate group, are displayedin red. Helices are depicted in white and &strands in yellow. The catalytic helix B runs in the plane of the page with N - to C-terminal ends running right to left. B: Rotation of 90" about helix B relative to that shown i n A .
change rates resulting in missing assignments or degeneracy and fewer structural restraints. The last 8 residues of the 174-residue domain are dynamically disordered, evident by their longer T2 values, and have been omitted from the model.
Stereochemical quaIity An evaluation using PROCHECK (Laskowski et al., 1993) reveals that the contacts,hydrogen bonding energies, and sidechain rotamers are typical of the quality of 2.5-A crystallographic structures or better. Peptide bond planarity and C a chirality are satisfactory aswell. The members of the ensemble on average have 87% of their non-glycine, non-proline residues in allowed regions of the Ramachandran map. On average, 9% of a member's residues are in marginally allowed regions of the Ramachandran plot and 4% are in disallowed regions. Wagner and coworkers have demonstrated that poor backbone dihedral angles in NMR solution structures cluster at residues whosetor-
sions are less reproducible across the ensemble, i.e., having lower backbone angular order parameters (cf. Hyberts et al., 1992). The best defied residues withangular order parameter in excess of 0.9(corresponding to a standard deviation for the torsion angle of ~ 2 5 ' )have conformationally orthodox backbone dihedral angles as shown in the Ramachandran plot for all 20 members of the ensemble (Fig. 6). The occurrences with positive 4 in the middle of the upper right quadrant belong to Gly 159 and Gly 161 (20 points each) of a calcium binding site.
Inhibitor binding The inhibitor runs antiparallel to strand IV (in Fig. 7 and Kinemages 1 and 2, topto bottom for "N- to C-") with an extended conformation, having medium to strong duNsequential NOES. The i-butyl side chain of the residue lacking an amide packs deeply (awayfrom the viewer) into the very deep and hydrophobic Si subsite where it (particularly its methyl groups) has
249 I
Solution s/ructure of inhibited strotnel.vsin
I
31
0 BACKBONE
+ALL
i
RESIDUE
talloproteinases recently reported. Reported locations include: in the active siteof a neighboring protease molecule ( Lovejoy et al., 3994b); in its own active site (Gooley et al., 1994); pointing away into "solution" (i.e., the crystal) (Bode et al., 1994; Borkakoti et al., 1994; Stams et al., 1994); and along thecarboxyterminal helix C (Lovejoy et al., 1994a; Reinemer et al., 1994). Like the latter two reports, the mature N-terminus (Phe 83) of the catalytic domain of stromelysin in this NMR model runs along the surface of helix C. Reinemer et al. (1994) reported that the amino groupof the N-terminal phenylalanine forms a salt bridge with the fully conserved Asp232 in neutrophil collagenase properly processed at the correct N-terminal residue to give full activity. This work suggests the disordered anddivergent locations of the N-termini in the structural models of MMPs is caused in part by the differences in N-terminal starting positions
NOEs to Leu 164, Val 198, His 201, Pro 221, Leu 222. and Tyr 223. The Pi leucine residue fits into the broad, shallow Si hydrophobic subsite at the surface where it has NOEs to Asn 162, Val 163, Leu 164, Ala 165, and Leu 222. The lack of stereospecific methyl assignments for the unlabeled inhibitor limits its precision in the model. The phenylalaninamide at P i occupies a shallow notch whereit has NOEs to Val 163, Leu 1 6 4 , Thr 190, Thr 191, Thr 193, Leu 222, Tyr 223, and His 224. Discussion
Terminal regions
The location of the amino-terminusvaries significantly among the crystallographic andNMR structural modelsof matrix me-
40 35
HEAW
Fig. 3. MeanRMSD (A) to the average structure among the 20 accepted structures of stromelysin catalytic domain. RMSDs for all heavy atoms of each residue are indicated by diamonds connectedwith lines. RMSDs for [he backboneheavy atoms ( N , CA, C', 0 )of each residueare shown using gray bars. Inhibitor positions occupy the last three positions (252-254) in the plot, following the break after protein residue 248.
I m LONG OMEDlUM OSEPUENTIAL
I , n
m
0
m
0
0
"
.
"
0 "
-
0 "
N
0 "
0
N
0
e
N
0
m
N
0
a
N
0
h
N
0
m
N
RESIDUE
0
m
"
0
N
0
.
. 0
n
,
e
: 0
m
U! . . .I>!.1 0
0
Fig. 4. Histogram of the number of interresidue N O E distance restraints used for each residue in the structure calculations. NOEs are separated into classes as longrange (dark), medium-range (medium). or sequential (light). The 50 NOEs discussed withreference to the first two rows of Table I are excluded from this tally. NOEs appear for both residuesinvolved. The inhibitorpositionsoccupythe last three positions in the plot.
S. R. Van Doren e[ al.
2492
A
' 0.9 0.8
0.7
K
a
0.6
0
8 PHI
0.3 0.2
0.1
0 O m
n O
n
n ~
"
n -
"
o N
"
"
n P
"
r 2
N
? 7
N
n C
N
r ,
N
? 0
N
n h
N
i
, m
r m
, O
n
r r
, N
n O
1 ?
O U
?
RESIDUE
B
1 0.9
Fig. 5. Histogram plotting angular order paramctcrs (Hyberts et al., 1992) for d ( A ) and $ (I%) backbone torsion angles. Standard deviations of a torsion angle of 10, 20.40, and 80" correspond to angular order parameters of 0.985, 0.942, 0.768, and 0.397, respectively.
0.8 0.7
a a
0.6
0
a a
0.5
8 PSI
0.3 0.2 0.1 0
of the domains used in such studies. The stromelysin domain used here has been found by N-terminal sequencing to begin at the "correct" residue. Because some of the members of this NMR ensemble havea distance of 5.5-7 A between Phe 83 and Asp 237 (without use of electrostatic terms in the restrained molecular dynamics to draw this closer in our model), it is quite plausible that properly processed stromelysin also has the same salt bridge between the amino-terminal phenylalanine and the equivalent aspartate. Role of metal
As stromelysin catalytic domain is most active at pH 6.0 (Ye molecule at et al., 1992), a logical choice would be to study the or near that pH, similar to the work reported by Gooley et at. (1994). However, we have observed that, at pH6.0, a large number of very sharp lines develops quickly at randomcoil proton chemical shifts in the NMR spectrum of inhibited stromelysin catalytic domain. The sharplines can be made to reversibly disappear by adjusting the pH to7.0 and by increasing the concen-
tration of Ca2+ ions in the buffer. These lines therefore cannot correspond to products of autolytic cleavage, but must represent unfolded areas of the protein. It was decided to study stromelysin at pH 7, in the presence of 20-fold excess of CaCI, to avoid the [local] unfolding. Stromelysin activity is just a factor of two lower than optimum at pH7 (Ye et al., 1992). We speculate that the unfolding at pH 6.0 occurs in the Ca2+ binding site regions because CaZ+ affinityis known to decrease 10-fold with a pH change from7 to 5.5 (Wilhelm et at., 1993). At least two calcium ions bind to stromelysin, as was determined by mass spectrometry (Hu et al., 1994). It is expected that they play an analogous role in stabilizing the 0-sheet as in collagenase, where X-ray diffraction studies identified clusters of acidic residues coordinating themetal ions in these regions (Lovejoy et al., 1994b). The current NMR studyreveals one such clusterof conserved acidic residues in stromelysin by the NOES of the side chain of Glu 184 with the side chains of both Asp158 and Asp 181. This cluster bridges the loop between strands 111 and IV with the loop at theC-terminal end of strandV. A second cluster of conserved carboxylate-containing residues, Asp 107 and Asp 182, are in
2493
Solution structure of inhibited stromelysin 180 150 120 90
.
60
30
w o -30
4
60
-90 -120 -150
-180 1 5 0 -120 -90
-60
-30
30
60
90
120 180 150
Fig. 6. Ramachandran plot for the residues of all 20 members of the ensemble whose angular order parameters for both 6 and exceed 0.9.
NOE contact and join the strand I to helix A loop with the strand V to helix B loop. These two respective clusters of negative charge are the probable binding sites for one Ca2+ion each as ahigh electrostatic energy cost would be imposed if calcium
were absent from each of these two sites. At the first calcium site inthe loop between strands 111and IV, NOESfrom the backbone of conserved Gly 159, conserved Gly 161, and Val 163 to either of the calcium-binding Asp 158 or Glu 184 residues are consistent with the crystallographic observation of three additional calcium-binding carbonyl groups. These are the carbonyl oxygens of the equivalent two glycines and asparagine of fibroblast collagenase (MMP-1; Lovejoy et al., 1994a). Becausethe six distances of calcium to ligand at this site, ranging from 1.8 to 2.5 A, reported by Lovejoy et al. (1994a) are consistent with the NOE data, they have been included inthis model. Some crystallographic models of collagenase include a third calcium ion just beyond the opposite edge of the sheet between the loops feeding into the N-terminal ends of @-strandsIll and V. A structuralzinc also stabilizes the @-sheet.A structuralzinc, in addition to the catalytic zinc ion, was found by Salowe et al. (1992). The two conserved histidines of strands IV and V, His 166 and His 179, adjacent and in NOE contact on the convex side of the sheet were implicated in ligation of this zinc (Gooley et al., 1993; Van Doren et al., 1993). From the loop joining strands I11 and IV a third histidine, His 151, wasalso found to coordinate this zincon the basis of its imidazole assignmentsand NOES (Gooley et al., 1993, 1994). Because Asp 153 is in NOE contact with His 151 and is fully conserved, it isthe best candidate to provide the fourth ligand to complete an ordinary tetrahedral coordination geometry. The recent X-ray structures have confirmed these observation, and the metal-ligand distances reported by Lovejoy et al. (1994a) have been included in
Fig. 7. Van der Waals surface plot of the average structure minimized with restraints displays the hydrophobic active site with bound drug fromall 20 structures colored red. The catalytic zinc ion (white sphere) is in the center and the structural calcium ion near the primed subsites is near the bottom (light blue sphere). A ribbon runs through the proteinC a coordinates. In order to highlight the hydrophobicity of the active site region, the hydrophobic protein residues are colored green and the hydrophilic residues are colored white and the glycine and tyrosine residues of intermediate hydrophobicity are colored light green. The superimposition used was an all-heavy atom superimposition for the entire protein-drug complex. In order to display the deep SI‘subsite, the hydroxamate residue has been omitted from the heavy atom plot of the ensemble of inhibitor conformations. Val 163 is obscured from view underneath the first two residues of the inhibitor shown.
S.R. Van Doren et al.
2494 this model. &Strand IV and the loopimmediately preceding, rigidified by the distal Zn2+ andproximal Ca2+, pack against the substrate analogue, particularly at the Si subsite in this model. Thus, thebridging by divalent cations around theperiphery of the sheet may be necessary to maintain the structure of the sheet and of one side of the active site cleft. The active site zinc coordination by three histidyl nitrogen ligands was expected because of the homology with the zinc endopeptidase astacin, whose structure was solved (Bode et al., 1992). This was verifiedfor collagenases bycrystallography and for stromelysin by diagnostic imidazole chemical shifts (Gooley et al., 1993). The fourth ligand in uninhibited matrix metalloproteinases is expected to be a water molecule able to carry out nucleophilic attack on thescissile carbonyl oncepolarized by a conserved glutamate in helix B acting as a general base, like a mechanism proposed for thermolysin by Matthews (1988). Hydroxamate-substituted inhibitors like the one used in this study must displace this water in providing the fourth and fifth ligands.
Comparison with other MMP structures The sequence similarity of stromelysin catalytic domain to crayfish astacin through three helices was pointed out earlier (Van Doren et al., 1993). The similarity of the tertiary structure of bacterial thermolysin with that of stromelysin catalytic domain has also been discussed by Gooley et al. (1994). In 1995, coordinates of a few matrix metdoproteinase structural models became available. The catalytic domains of human fibroblast collagenase and human stromelysin have 61% sequence identity when using the alignment of Whitham et al. (1986) and can be expected to have quite similar tertiary structures. A superimposition with the fibroblast collagenase crystallographic model of Lovejoy et al. (1994b; Protein Data Bank access code ICGE) appears in Figure 8 and Kinemage 2. This figure shows that the
catalytic domains of these two respective matrix metalloproteinases superimpose quite well throughout theP-sheet and helices. When omitting the first eight residues of this NMR model of stromelysin (and the fiist six residues of the collagenase) because of the influence of conditions on the position of the collagenase N-terminus as discussed above, the pairwise RMSD for the backbone heavy atoms (N, CA, C‘, 0)is 2.70 A. (Structures were overlaid using Biosym’s Insight I1 and RMSD calculated using the routine of Chris Ingalls.) The loop connecting P-strands 111 and IV superimpose poorly, partly because of a translation in the region of the calcium binding loop (see below). A threeresidue insertion in stromelysin (Whitham et al., 1986) occurs in the long loop of irregular structure between helicesB and C: stromelysin: 229 T D L T R F R S L Q D D 238 collagenase:
G D V
QLAQ D D
When omitting 14 residues of the loop joining strands I11 and IV and 3 residues on each side of the insertion, the pairwise backbone RMSD improves to 2.28 A. The solution structure of stromelysin’s catalytic domain reported by Gooley et al. (1994) wasdetermined under conditions of lower calcium affinity (pH 5.5; see discussion above) and of lower calcium concentration ( 5 mM). The average backbone RMSD between five arbitrarily chosen members of the ensemble of Gooley et al. (1994) and five arbitrarily chosen members of our ensemble is 1.87 0.3 A. For comparison, the average pairwise backbone RMSD of one member of our ensemble from another member is about 1.3 A. The superimposition of the structure of Gooley et al. and ours is also displayed in Figure 8. Most of the disparities between the model reported here and the models of Lovejoy etal. (1994b) and Gooley et al. (1994) occur in loops. In the solution structures, the loops are generally more poorly defined (Figs. 2, 3, 5) and more poorly re-
*
Fig. 8. Overlay of the current restrained average structureof stromelysin catalytic domain(yellow) with the X-ray crystallocode ICGE,white) and the graphic structureof collagenasecatalytic domain (Lovejoy et al., 1994b; Protein Data Bank access N M R structure of stromelysin catalytic domainas determined by Gooley etal. (1994; 1995 updated structure Protein Data Bank access code ZSRT, light-blue).
Solution structure of inhibited stromelysin
2495
protection of the amide protons of Tyr 223 and Ala 165 on strained with NOEs (Fig. 4). The high degree of solvent exposure opposite rims of the active site cleft even though these groups is largely responsible for this limitation on the number of NOE are not involvedin canonical secondary structure hydrogen restraints. The solvent exposure limits the number of assignof the bonding patterns. Regrettably, the present definition ments because of higher amide proton exchange rates andbestructure is not sufficient toclearly identify the hydrogen bond cause less groups are packed to yield NOEs. Mobility of the acceptors that could be in the substrate analogue. exposed loops may also reduce NOE intensity. The difficulty in Starting at the “southwest” end at thePi position, the pheobtaining complete aromatic assignments, particularly of phenylalanines, in proteins of thissize is also a limitation. Such fac- nylalaninamide occupies a nook so shallow (Si), it is questionable whether it is a subsite of any impact upon specificity or tors explain the limited number of NOEs in residues 149-156 of affinity. Indeed, the substratespecificity studies of Niedzwiecki the loop joining 0-strands 111 and IV, for example, and thus et al. (1992) revealed a need for noresidue, but perhaps a small probably compromise the accuracy as well as the precision in hydrogen bond donor at the Pi position. Figure 7 would sugunderdetermined loop regions such as this. The calcium bindgest Si is somewhat hydrophilic. The Pi position of the inhibiing region of the loop between strands 111 and 1V is, however, well defined in our solutionmodel of stromelysin. This loop ap- tor (phenylalaninamide) is quite well defined despite the2.4-A pseudoatom corrections in the NOE restraints, because of the pears as alarge S-shaped structure in the top right-handhalf of large number of the restraints. Thisis surprising because of the Figure 8. Yet, a large difference exists in the position of this loop in our structure as compared to both the structures of Lovejoylack of a pocket at the S i subsite. The aromatic resonancesof this Pi phenylalaninamide are remarkably sharp,having an inet al. (1994b) and Gooley et al. (1994). This difference is a transtrinsic line width of 8-10 Hz when subtracting away a 7-Hz conlation of 4-5 A, whereas the conformationof the loop itself is approximately the samein the structures. In particular, the loop tribution from scalar coupling.Most proton resonances of the in our structureis much more closely packed against the inhib- protein are much broader,having widths of at least 20 Hz. This suggests the phenylalaninamideresidue may actually experience itor and against the loop in the region of residue number 190 a dynamic equilibrium between a conformationin which it packs (center right in Fig. 8). We list here several key NOEs that seem against the protein andtransiently builds up a numberof NOEs to dictate this difference:these are NOEs Asn 162 Ha-Gly 192 and a statein which it is more freely moving and exposed to the HN and Val 163 methyl-Thr 193 H a , which are completely unsolvent. Thus, the high precision of Pi is likely to be an artifact ambiguous by themselves and which are supported by several of conventional structure calculation methods. Such dynamic other NOEs (e.g., 161-193 and 160-193). In our average strucbehavior could possibly be better modeled with long trajectory ture, the distancesbetween the protons involved are 5.4 A and molecular dynamics simulationsusing time-averaged NMR re5 . 2 A for the 162-192 and 163-193 NOEs, respectively. In constraints, keeping in mind that the detailsof the dynamic model trast, the structurereported by Gooley et al. (1994) yields 16 and derived by such methods may be of limited value. Such an ap12 A for these distances. We must therefore conclude that siga proach has accounted for discrepancies between high-resolution nificant and major differenceexists between the packing of this NMR and crystallographic modelsof chymotrypsin inhibitor 2 structural Zn/Ca loop against the protein in the different reas the differing conformationssatisfied restraints when averagported protein structures. ing over time (Nanzer et al., 1994). We d o not have any indication that other residues of the bound inhibitor are subject to Bound substrate analogue similar averaging: their resonances are broad, and their interThough several stereoscopic views of inhibitors bound to fibro- molecular NOEs have similarintensities as protein intramolecblast and neutrophil collagenases have been presented and some a ular NOEs. The tightly bound state for the inhibitor as whole coordinates of the complexes havebeen more recently released inferred from these observations is in accordance with the mea(Bode et al., 1994; Borkakoti et al., 1994; Lovejoy et al., 1994a; sured macroscopic K, of IO-’-lO-* M (depending on substrate) Stams et al., 1994), only a single set of coordinates of a drug at the solution conditions employed in this study (15% acetobound to stromelysin was available at the time of this writing. nitrile) (Q.-Z. Ye, unpubl.). Recently determined structures show thatwhen a substrate anThe S; subsite occupied by the leucine side chain is quite hyalogue is bound, it runs antiparallel to 0-strand IV at the edge drophobic though open to the surface. The lack of definition of the sheet. Most of the complexes, including the one reported of the P; leucine in the subsite is a consequence of the comparhere, have the substrate bound in the primed sites (C-terminal atively limited number of NOE assignments and of the large to the scissile bond) in an extended fashion. A hydroxylamine- methyl-methyl pseudoatom correction in the restraints. The Si substituted tripeptide complexed with neutrophil collagenase, subsite appears tobe a shelf broad enough to accommodatevery however, binds to the unprimedsites, mimicking the N-terminal bulky hydrophobic side chains. This agrees with the marked end relative to thescissile bond ofa substrate (Bodeet al., 1994). preference for phenylalanineor tryptophan at the Si subsite in A substrate then canbe expected to drape across the active site both inhibition studies (Darlak et al., 1990; Kortylewicz & of a matrix metalloproteinasein an extended fashion. The por- Galardy, 1990; Grobelny et al., 1992) and substrate specificity tion of the stromelysin active site that a natural substrate would studies of homologouscollagenase (Netzel-Arnett et al., 1991). occupy should be from [N- to C-terminus] “northeast to south- Limited studies in stromelysin reveal a similar preference for west” in Figure 7. This swath is very hydrophobic, asrepresented tryptophan at S; as a determinant of specificity in inhibition by the green color of the Van der Waals surface. A significant (Ye et al., 1994) and in substrate hydrolysis (Niedzwiecki etal., fraction of substrate binding energy is therefore expected to arise 1992), again consistent with the site seen in Figure 7. from the interaction of the substrate hydrophobic side chains Thorough studies of stromelysin’s Si subsite specificity rewith this surface. That there may also be electrostatic contribuquirement show a clear preference for hydrophobic residues, tions to the binding energy is suggested by the exchangeparticularly phenylalanine and leucine (Niedzwiecki et al., 1992).
S. R. Van Doren et al.
2496 They found the unbranched, unnatural amino acid norvaline in this site to be threefold more active than these. These observations are consistent with the depth and narrowness of the hydrophobic Si site seen in Figure 7. If “C labeling of the inhibitor had been available to enable stereospecific assignments of its methyl groups, the definitionof the Pi i-butyl side chain in this deep pocket would still be better. The depthof the Si site has been reported in the crystallographic models of collagenase as well. Thus, the specificity studies qualitatively agree with the structural features of this stromelysin-inhibitor complex and may aid drug design. We have compared the binding mode of the inhibitor in the current structure with that in the recently released stromelysin structure reported by Gooley et al. (1994). In both cases, the inhibitor binds in an extended conformation. The location ofthe Pi side chain and the location of theSi pocket are identicalin both structures. However, the position of the inhibitor backbone is moved further to the “front”of the molecule (in the view of Fig. 8) in our structure. This places the loop around Pro 221 (the lower “lip” of the binding cleft,see Fig. 8) significantly farther forward and lower. Although i t is very difficult to distinguish cause and effect of conformational differences between structures, we note that theP2‘ leucyl side chain in our molecule, being hydrophobic, is more tucked into the S2‘ pocket than the arginyl side chain in the model reported by Gooley et al. This placement could cause the shift of the inhibitor forward. Alternatively, the large significant shift in the position of the structural Zn/Ca loop (see above) could also cause this change of active site, or, conversely, this particular inhibitormay induce the change in conformation of that loop. Not neglected as possible causes of the structural differences canbe the differences in pH, Ca2+ion concentration, and the presence of acetonitrile as a co-solvent. We, however, conclude that the structural differences cannot be dismissed as being caused by indetermination of the areas involved. Nor can they be easily brushed away by challenging the NOE data our structureis based on: several unambiguous NOEs define the position of the areas involved. Differences between the structuresmight thus reflect differences in interactions between the structural Zn/Ca loop and the active site cleft at different solvent conditions. Methods N M R data collection and handling
The stromelysin catalytic domainwas expressed in Escherichia coli and purified frominclusion bodies as describedpreviously by Ye et al. (1992). Preparation of the labeled samples, typically about 1 mM, and of the buffers (containing I O rnM Tris-dllHCI, pH 7.0, 20 mM CaCI,, 15% acetonitrile43 and 8% D,O) was as described by Van Doren et al. (1993). The hydroxamatesubstituted inhibitor (IC1 U24522; DiPasquale et al., 1986) was present at a ratioof 1: 1 with the protein for thehalf-filtered experiments (cf. Ikura & Bax, 1992). In all other experiments, the inhibitor concentration exceeded the protein concentration. Thecross-polarization-drivenHCCH-TOCSY (HEHOHE HAHA;Majurndar et al.,1993),HSQC(Bodenhausen & Ruben, 1980), HCH (Yamazaki et al., 1993), HMQC-J (Kay& Bax, 1990), and gradient-enhanced NOESY-HSQC and HSQCNOESY (Majumdar & Zuiderweg, 1993) spectra were acquired with an AMXSOO spectrometer (Bruker Analytische Messtech-
nik GMBH, Karlsruhe, Germany) equipped with a Bruker Grasp unit and triple resonance gradient probe. The CBCA(C0)NH (Grzesiek & Bax, 1992), HBHA(C0)NH (Grzesiek & Bax, 1993), ”C-resolved FSCT-HSMQC-NOESY (Van Doren & Zuiderweg, 1994), half-filtered NOESY, and filtered TOCSY(Bax et al., 1994) spectra were acquired on an AMX-600 equipped with an auxiliary fourth channel (Van Doren & Zuiderweg, 1993). Spectra were processed using Felix (Hare Research, Inc., Bothell, Washington) and interpretedusing the Felixtalk interface for Felix (written by Dr. A. Majumdar). Restraints
Interresidue NOE restraints involving amides were obtained from 3D [“NINOESY-HSQC (Fesik & Zuiderweg, 1988; Marion et al., 1989; with modifications of Van Doren et al., 1993) and 3D [”CIHSQC-NOESY, both collected in H,O using gradient enhancements (Majumdar & Zuiderweg, 1993) at 5 0 0 MHz with 70 ms mixing time. 3D [13C]FSCT-HSMQC-NOESY(Van Doren & Zuiderweg, 1994) in D,O at 600 MHz with 55 ms T,,,,carrier at 76 ppm, and SUSAN-I broadband decoupling scheme (Sunitha Bai et al., 1994) was the principle source of NOE restraints amongboth aliphatic and aromatic groups.Assignments of the peptide-derived inhibitor relied on a TOCSY(35 msmixing time) with “C-bound protein protons suppressed using J cross-polarization (Bax et al., 1994) and on [F,-”C, F2-”C] NOESY with I3C-bound protonssuppressed in both dimensions (Ikura & Bax, 1992), both recorded at 600 MHz using uniformly “C/”N-labeledstromelysincomplexedwithanunlabeled hydroxamate-substitutedpeptide-likeinhibitor in DzO. At 600 MHz, 2D “C-suppressing [FI-”C] NOESY in D,O and I3C/”N-suppressing [Fz-12C/’4N] NOESY in H,O (cf. Ikura & Bax, 1992), each with 100 ms mixing times, revealed NOEs between inhibitor and protein. An HMQC-J (Kay & Bax, 1990) provided @ angle restraints. NOE distance restraints with 1.8-A lower bounds were conservatively assigned upper bounds of 2.7 A for very strong, 3.2 A for strong, 4A for medium, or 5 A forvery weak NOEs. Participation of methyl groups pushed the restraint into the next larger bin. Conventional pseudoatom corrections (Wuthrich, 1986) were applied. Fifty-five @ torsion angles were loosely restrained using ranges of -160 to -80 for ‘ J H n H C r > 9 HZ,-170 to -70 for ’JHnHlu > 8 Hz, -180 to -60 for ? J H ” H < , 8 Hz provided d r r N ( i , i not ) strong, or -90 to -30 for 3 J H n H , r < 6 Hz, similar to ranges used by Clubb et al. (1994). Forty-two hydrogen bonds (two restraints per bond) were included for slowly exchanging amides (present after 18 h at pH 7 at 32 “C) in the sheet and helices. The bounds for such amide protons to carbonyl oxygens were 1.5 and 2.3 A , whereas the bounds for such amide nitrogens tooxygen were 2.5 and 3.3 A in order to maintain linearity. Distance restraints between two zinc atoms and their histidyl nitrogen ligands (bounds of 1.95 and 2.25 A), consistent with boththeimidazolenitrogenassignments of Gooley et al. (1993, 1994) and with the ligands seen in the crystallographic structure of the homologous fibroblast collagenase crystallographic structure (Lovejoy, 1994a), were employed. Six calcium to oxygen distance restraints (to Asp158 061, Asp 181 062, Glu 184 0 6 2 with bounds of 1.75 and 2.25 A ; to Gly 159 0, Gly 161 0, Val 163 0 with bounds of 2.05 and 2.55 A ) were used in the modelbased on the structureof the homologouscollagenase (Lovejoy et al., 1994a). The hydroxamate substituent
-
Solution structureof inhibited strornelysin
at the N-terminus of the peptidic inhibitor was modeled with a distance restraint (1.75 and 2.25 A)from the catalyticZnZ+to each of its oxygens in accordance with the crystallographic results for collagenases with hydroxamate inhibitors bound (e.g., Borkakoti et al., 1994; Stams et al., 1994). Omega torsion angle restraints were added to maintaintrans planarity of the peptide bonds of non-proline residues.
2491
Hansen B. 1993. Matrix metalloproteinases: A review. Crii Rev Oral Biol Med 4:197-250. Bode W, Gomis-Ruth FX, Huber R, Zwilling R, StockerW. 1992. Structure of astacin and implications for activation astacins and zinc-ligation of collagenases. Nulure 358:164-167. Bode W, Reinemer P, Huber R, Kleine T, Schnierer S, Tschesche H. 1994. of human neutrophil The X-ray crystal structure of the catalytic domain collagenase inhibited by a substrate analogue reveals the essentials for catalysis and specificity. EMBO J 13:1263-1269. Bodenhausen G, Ruben DJ. 1980. Natural abundance nitrogen-I5 NMR by enhanced heteronuclear spectroscopy. Chem Phys Leti 69:185-189. Borkakoti N, Winkler FK, Williams DH, D'Arcy A, Broadhurst MJ, Brown Strucrure calculations and evaluation RA, Johnson WH, Murray EJ. 1994. Structure of the catalytic domain of human fibroblast collagenase complexed with an inhibitor. Naiure Distance geometry was used to generate 39 starting structures Siruct BIOI 1:106-1 IO. then optimized by distance-driven dynamics (simulated annealClavel C, Polette M, Doc0 M, Benninger I , Berembaut P. 1992. Immunoing without a physical forcefield) using DGll (Havel, 1991; Biolocalization of matrix metalloproteinases and their tissue inhibitors in sym Technologies, Inc., San Diego, California). Using Discover human and mammary pathology. Bull Cancer 79:261-270. Clubb RT, Ferguson SB, Walsh CT, Wagner G. 1994. Three-dimensional so(Biosym), the 39 structures were further refined using a simulution structure ofEscherichia coli periplasmic cyclophilin. Biochemislated annealing protocol using the Amber forcefield without use iry 33:2761-2772. of charges, followed by extensive restrained minimization. The Darlak K , Miller RB, Stack MS. Spatola AF, Gray RD. 1990. Thiol-based inhibitors of mammalian collagenase. J Biol Chem 265:5199-5205. 10 structures having the highest energy (sum of restraining and DiPasquale G, Caccese R, PasternakJ , Conaty J , Hubbs S, Perry K . 1986. conformational terms) were removed from the ensemble. These Proteoglycan- and collagen-degrading enzymes from human interleuken-I I O have some of the highest RMSDs from the other members stimulated chondrocytes from several species; proteoglycanase and colof the ensemble, judging from pairwise RMSDs. Two more lagenase inhibitors as potentially new disease-modifying antiarthritic agents. Proc Soc Exp Biol Med 183:262-267. structures were culled on the basis of single large violations Fesik SW, Zuiderweg ERP. 1988. Heteronuclear three dimensional NMR greater than 0.75A . Another seven structures were culled on the spectroscopy; a strategy for the simplification of homonuclear two dimenbasis of unique, localized aberrations or conformational stresses sional NMR spectra. J Magn Reson 78:588-593. Firestein GS, Paine MM. 1992. Stromelysin and tissue inhibitor of metalloin regions well defined in the other structures andwere accomproteinases gene expressionin rheumatoid arthritis synovium. Am J Papanied by an unusually large number of violations in these re/hol 140:1309-1314. gions. Twenty of the 39 were accepted. These structures were Gooley PR, Johnson BA, Marcy AI, Cuca GC, Salowe SP, Hagman WK, superimposed using all backbone heavy atoms or all heavy atEsser OK, Springer JP. 1993. Secondary structure and zinc ligation of human recombinant short-form stromelysin by multidimensional heterooms using Insight11 (Biosym). Theaverage structure was refined nuclear NMR. Biochemisiry 32:13908-13108. by extensive restrained minimization. Mean RMSD values to the Gooley PR, O'Connell JF, Marcy AI, Cuca GC, Salowe SP, Springer JB, average structure, residue by residue, were tabulated using softJohnson BA. 1994. The NMR structure of the inhibited catalytic domain of human stromelysin-I. Nature Struct Biol 1:111-118. ware written by Chris lngalls of Parke-Davis/Warner-Lambert Gravallese EM, Darling JM, Ladd AL, Katz JN, Glimcher LH. 1991. In situ Co. Angular order parameters (Hyberts et al., 1992) were calhybridization studies of stromelysin and collagenasemessenger RNA exculated using routines provided by Dr. Sven Hyberts and aninpression in rheumatoid synovium. Arthritis Rheum 34:1076-1084. Grobelny D, Poncz L, Galardy RE. 1992. Inhibition of human skin fibroterface written by Chris Ingalls. blast collagenase, thermolysin, and Pseudomonas aeruginosa elastase by peptide hydroxamic acids. Biochemistry 31:7152-7154. Gronenborn AM, Clore GM.1994. Identification of N-terminal helix capCoordinates ping boxes by means of "C chemical shifts. J Biomol NMR 4:455-458. Grzesiek S,Bax A. 1992. Correlating backbone amide and side chain resoThe atomic coordinates have been deposited with the Protein nances in larger proteins by multiple relayed triple resonance NMR. J Data Bank. The ID codes arelUMS (an ensemble of 20 strucA m Chem Soc 114:6291-6293. tures of stromelysin catalytic domain complexed with hydroxGrzesiek S, Bax A. 1993. Amino acid type determination in the sequential assignment procedure of uniformly I3C/"N-enriched proteins. J Bioamate inhibitor IC1 U24522) and IUMT (the energy minimized mol NMR 3:185-204. average structure). Havel TF. 1991. An evaluation of computational strategies for use in the determination of protein structure from distance constraints obtained by nuclear magnetic resonance. Prog Biophys Mol Biol 56:43-78. Acknowledgments Hu P, Ye QZ, Loo J A . 1994. Calcium stoichiometry determination for calcium binding proteins by electrospray ionization mass spectrometry. Anal T h i s w o r k w a s s u p p o r t e d b y N I H g r a n t G M R 52406-01 01 to E.R.P.Z Chem 66:4190-4194. and by American Cancer Society fellowship PF-4056t o S.R.V. We thank Hyberts SG, GoldbergMS. Havel TF, Wagner G. 1992. The solution strucDrs. C.C. Humblet, M.D. Reily, and G.D. Glick for lending use of ture of eglin c based on measurements of many NOES and coupling concomputer time for calculation of 24 of the starting structures in the time- stants and its comparison with X-ray structures. Protein Sci /:736-751. consuming distance geometry step. We thank Mr. C. Ingalls for conlkura M, Bax A. 1992. Isotope-filtered 2D NMR of a protein-peptide comtributing RMSD analysis software and Dr.S.V. Hyberts for contributing plex study of a skeletal muscle myosin light chain kinase fragment bound scripts for measurement of angular order parameters. We acknowledge to calmodulin. J A m Chem Soc~114:2433-2440. M.W.F. Fischer and H. Wang for software assistance. We appreciate Kay LE, Bax A. 1990. New methods for the measurement of NH-CaH coustimulatingdiscussionswithDrs.H.Nagase, V. T h a n a b a l , a n d A . pling constants in 15N-labeled proteins. J Magn Reson 86:llO-126. Majumdar. Kortylewicz ZP, Galardy RE. 1990. Phosphoramidate peptide inhibitors of human skin fibroblast collagenase. J Med Chem 33:263-273. Kraulis P. 1991. MOLSCRIPT: A program to produce both detailed and schematic plots of protein structures. ./ Appl Crystallogr 24:946-950. References LaskowskiRA,MacArthurMW,MossDS,ThorntonJM.1993.PROBax A, Grzesiek S, Gronenborn AM, Clore GM.1994. Isotope-filtered 2D CHECK: A program to check the stereochemical quality of protein strucHOHAHA spectroscopy of a peptide-protein complex using heteronutures. J Appl Crystullogr 26:283-291. clear Hartmann-Hahn dephasing. J Magn Reson A 106:269-273. Li J, Brick P, O'Hare MC, Skarzynski T, Lloyd LF, Curry VA, Clark IM, Birkcdal-Hansen H,MooreWGI,BoddenMK,WindsorLJ,BirkedalBigg HF, Hazleman BL. 1995. Structure of full-length porcine synovial
2498 collagenase reveals a C-terminal domain containing a calcium-linked, four-bladed &propeller. Structure 3:541-549. Lovejoy B, Cleasby A, Hassell AM, Longley K, Luther MA, Weigl D, McGeehan G, McElroy AB, Drewry D, Lambert MH, Jordan SR. 1994a. Structure of the catalytic domain of fibroblast collagenase complexed with an inhibitor. Science 263:375-377. Lovejoy B, Hassell AM, Luther MA, Weigl D, Jordan SR. 1994b. Crystal structures of recombinant 19-kDa human fibroblast collagenase complexed to itself. Biochemistry 33:8207-8217. Majumdar A, Wang H, Morshauser RC, Zuiderweg ERP. 1993. Sensitivity improvement in 2D and 3D HCCH spectroscopy using heteronuclear cross-polarization. J Biomol N M R 3:387-397. MajumdarA,ZuiderwegERP.1993.Improved‘3C-resolvedHSQCNOESY spectra in H 2 0 , using pulsed field gradients. J Magn Reson B 102:242-244. Marion D, Kay LE, Sparks SW, Torchia DA, Bax A. 1989. Three-dimensional heteronuclear NMR of 15N-labeled proteins.J A m Chem Soc 111:15151517. Matrisian LM. 1992. The matrix-degrading metalloproteinases. BioEssuys 14:455-463. Matthews B. 1988. Structural basis of the action of thermolysin and related zinc peptidases. Ace Chem Res 21:333-340. Nagase H . 1995. Matrix metalloproteinases. I n : Hooper NM, ed.Zincmetalloproleases in health and disease. Chichester, UK: Ellis Horwood. Nanzer AP, Poulsen FP, van Gunsteren WF, Torda AE. 1994. A reassessment of the structure of chymotrypsin inhibitor 2 (CI-2) using timeaveraged NMR restraints. Biochemislry 33:14503-1451 I . Neri D, Szyperski T, Otting G , Senn H , Wiithrich K . 1989. Stereospecific nuclear magnetic resonance assignmentsof the methyl groups of valine and leucine in the DNA-binding domain of the 434 repressor by biosynthetically directed fractional I3C labeling. Biochemistry 28:7510-7516. Netzel-Arnett S, Fields G, Birkedal-Hansen H, Van Wart HE. 1991. Sequence specificities of human fibroblast and neutrophilcollagenases. J Biol Chern 266:6747-6755. Niedzwiecki L, Teahan J, Harrison RK, Stein RL. 1992. Substrate specificity of the human matrix metalloproteinase stromelysin and the development of continuous fluorometricassays. Biochemistry31:12618-12623. Okada Y, Shinmei M , Tanaka 0, Naka K, Kimura A, Nakanishi I , Bayliss MT, lwata K , Nagase H. 1992. Localization o f matrix metalloproteinase 3 (stromelysin) in orteoarthritic cartilage and synovium. Lab Invest 66580-690. Okada Y, Takeuchi N, Tomita K , Nakanishi 1, Nagase H . 1989. Immunolocalization of matrix metalloproteinase-3 (stromelysin) in rheumatoid arthritis. Ann Rheum DIS48:645-653. Reinemer P, Grams F, Huber R, Kleine T, Schnierer S , Piper M, Tschesche H, Bode W. 1994. Structural implications for the role of the N terminus in the “superactivation” of collagenases. FEBS Lett 338:227-233. Salowe SP, Marcy AI, Guca GC, Smith CK, Kopka IE, Hagmann WK, HermesJD. 1992.Characterizationofzinc-bindingsitesinhuman stromelysin-I: Stoichiometry of the catalytic domain and identification of a cysteine ligand in the proenzyme. Biochemistry 3/:4535-4540. Sasaguri Y, Komiya S , Sugama K, Suzuki K, lnoue A, Morimatsu M, Nagase H. 1992. Production of matrix metalloproteinases 2 and 3 (strome-
S.R. Van Doren et al. lysin) by stromal cells of giant cell tumor of bone. A m J Palhol 141: 61 1-621. Shima 1, Sasaguri Y, Kusukawa J, Yamaha H, Husita H, Hakeyawa J, of matrixmetalloproteinase-2and MorimatsuM.1992.Production metalloproteinase-3 related to malignant behavior of esophageal carcinoma. Cancer 70:2747-2753. Stams T, Spurlino JC, Smith DL,Wahl R C , H oTF, Qoronfleh MW, Banks TM, Rubin B. 1994. Structure of human neutrophil collagenase reveals large Sl‘specificity pocket. Nature Struct Biol 1:119-123. Stetler-Stevenson WG, Aznovoorian S, Liotta LA. 1993. Tumor cell interAnnu actions with the extracellular matrix during invasion and metastasis. Rev Cell Biol9:541-573. Sunitha Bai N, Kari N, Ramachandran R. 1994. An efficient windowless seJ Magn Reson A quence for broadband heteronuclear decoupling. 106:241-244. Van Doren SR, Kurochkin AV, Ye QZ, Johnson LL, Hupe DJ, Zuiderweg ERP. 1993. Assignments for the main-chain nuclear magnetic resonances and delineation of the secondary structure of the catalytic domain of human stromelysin-l as obtained from triple-resonance 3D NMR experiments. Biochernistry32:13109-13122. Van Doren SR, Zuiderweg ERP. 1993. An auxiliary RF channel with convenient phase control for NMR spectrometers. J Mugn Reson A 104: 222-225. Van Doren SR, Zuiderweg ERP. 1994. Improvements in HSMQC-type double- and triple-resonance NMR experiments by using full-sweep (semi)constant-time shift labeling. J Magn Reson B 104:193-198. Walakovitr LA, Moore VL, Bhardwaj N, GallickGS, Lark MW. 1992. Detection of stromelysin and collagenasein synovial fluid from patients with rheumatoid arthritis and posttraumatic knee injury. Arthritis Rheum 35:35-42. Wilhelm SM, Shao ZH, Housley TJ, Seperack PK, Baumann AP, GunjaSmith Z, Woessner JF Jr. 1993. Matrix metalloproteinase-3 (stromelysinI ) : Identification as the cartilage acid metalloprotease and effect of pH on catalytic properties and calcium affinity. J Biol Chem 268:2190621913. Whitham SE, Murphy G, Angel P, Rahmsdorf HJ, Smith BJ, Lyons A, Harris TJR, Reynolds JJ, Herrlich P, Docherty AJP. 1986. Comparison of human stromelysin and collagenaseby cloning and sequenceanalysis. Blochem J 240:913-916. Woessner JF Jr. 1991. Matrix metalloproteinases and their inhibitors in connective tissue remodeling. FASEB J5:2145-2154. Wuthrich K. 1986. N M R of proteins and nucleic acids. New York: John Wiley & Sons. Yamazaki T, Yoshida M, Nagayama K. 1993. Complete assignments of magnetic resonancesof ribonuclease H fromEscherichia coli by double- and triple-resonance2Dand3DNMRspectroscopies. Biochemistry 32: 5656-5669. Ye QZ, Johnson LL, Hupe DJ, Baragi V. 1992. Purification and characterization of the human stromelysin catalytic domain expressed in E. coli. Biochemistry3l:l 1231-11235. Ye QZ, Johnson LL, Nordan I , Hupe D, Hupe L. 1994. A recombinant human stromelysin catalytic domain identifies tryptophan derivatives as human stromelysin inhibitors. J Med Chem 37:206-209.
